Stage 0: extract pure DSP (tested) + split dashboard.js by feature

[astrobyte-dev]

Stage 0: foundation — pure DSP extracted + tested, dashboard.js split by feature, selftest verifies audio

Closes all three Stage 0 gaps. Two commits:

• d16633a — GAP 1 (extract + test pure DSP) and GAP 2 (split dashboard.js)
• a79a2e5 — GAP 3 (selftest asserts master-meter activity)

What changed

• GAP 1: stabilize / resample / audioPhase extracted into a single-source dashboard-dsp.js (window.AbxDsp, globalThis.AbxDsp under Node). One body — the browser and the new src/dsp.test.ts import the same file, so there is no TS copy to drift; dashboard-dsp.d.ts is the type contract. audioPhase purified to (clk, nowMs). 12 new tests; suite is 19/19.
• GAP 2: the 672-line dashboard.js split into a core shell + five feature modules: AbxScope, AbxCheats, AbxCurves, AbxSeq, and a shared Abx core surface (state/send/cmd/poll/color/esc/fnum/clock.phase). dashboard.ts serves any /dashboard-*.js. Core 672 → 334; every feature module < 200.
• GAP 3: selftest.ts now binds a Meter on the master-meter UDP port, samples L/R across the 8s phase-B beat, and asserts peak(L+R) > 0.05; on failure it dumps sclang + tidal tails and exits non-zero. Local-only (npm run selftest), never CI.

---

Standing review

1. Karpathy / process. Design alternatives were surfaced before code (the namespace gate presented both the literal four-file split and the core+features cut with concrete line budgets; you chose the latter). Each of the six steps was a surgical, behaviour-preserving move, verified before the next.

2. Tests. 12 new cases for the pure DSP, behaviour-focused (sine locks to ~2 cycles vs kick/noise return raw; resample endpoints+interp+degenerate; audioPhase null/fractional/advance/lead-compensation). npm test ran green before and after every step. Single source means no JS/TS drift. GAP 3 adds the live audio gate (local-only) so a silently dead chain fails loudly.

3. Architecture. Core is 334 lines — the irreducible shell (render loop, the four dispatchers, master meters, transport, console). Intentionally over the 200 soft cap, as agreed; chasing sub-200 would mean shredding it into ~8 modules, a separate effort. Feature modules: dsp 28, scope 47, cheats 149, curves 91, seq 134. No duplication.

4. Security. No new user-controllable strings reach SC/Tidal. The new static-file route is regex-restricted (/^\/(dashboard(?:-[a-z0-9-]+)?\.js)/i — lowercase alnum + dash, ends .js, no slashes/dots), so it cannot path-traverse. Host/Origin guards untouched. No new persisted state.

5. Concurrency / lifecycle. No new long-lived timers: the existing 80ms debounce and rAF loops are reused; AbxSeq owns its row/swing queue, drained via flushPending() from the same interval. Each feature self-registers its own listeners. The core render loop guards window.AbxSeq/AbxCurves so it can start before the later scripts parse and survives a module failing to load. selftest binds + stop()s its own meter socket (unref'd).

6. Performance. Render-loop work is delegated, not added; guards are cheap truthy checks. No new poll endpoints or SSE streams.

7. Comments. Hard-won fixes carry the WHY: the load-race guard rationale, the noise-seed choice (a low-bit LCG faked a tone — mulberry32 instead), the DOMContentLoaded deferral, the clock.phase purity note, and the selftest threshold rationale (why 0.05).

8. UX regressions. Dispatch branches are mutually exclusive (distinct data-attrs), so reordering them into delegates changes nothing. Listener-order note: the curve-paint vs seq-paint pointerdown registration order is restored to the original (curves loads before seq), and the two act on mutually-exclusive targets (.curvecv vs .pad), so order is immaterial regardless. No keyboard shortcut or debounce threshold changed.

9. README / docs. No user-facing features added (pure refactor + a test), so the feature list, env-var table, and MCP tool list are unchanged.

10. Next-stage readiness. Stage 1 (mixer) can reuse AbxScope.draw (exposed) for mini-scopes, Abx.color for per-channel colour, Abx.state().scopes for meters, and the AbxX.handleClick delegate pattern. Harder than expected: the inter-script load race affected the core loop's calls into later-loaded namespaces, not just the deep-link — now guarded. No known debt carried into Stage 1.

---

Manual test checklist

Verified headlessly (Edge + puppeteer-core, 3 consecutive runs, zero console errors):

• Dashboard loads with no console/page errors
• Step grid / cheats / curves / samples panels open
• #steps and #cheats deep-links open via the new DOMContentLoaded bootstrap

Needs your ears + hands (audio + drag gestures I can't drive headlessly):

• Boot, play a beat → per-card scopes render
• Open step grid, paint a row → hear it; switch A→B→C; toggle song mode
• Open curves, paint a curve, toggle ON → hear modulation, OFF → stops
• Drag a sample onto a sequencer row → sound changes; drag a grip → row reorders
• Scroll a lit pad → velocity (brightness) changes; right-click → erase
• hush / record / reset behave as before
• npm run selftest (standalone — kills any live engine, makes ~8s of audio) prints DONE — audio verified.

[Copilot]

Pull request overview

This PR is a Stage 0 refactor that (1) extracts the dashboard’s DSP math into a single shared JS implementation with Node unit tests, and (2) splits the previously-monolithic dashboard UI script into a small core plus feature modules that consume a shared window.Abx surface. It also updates the dashboard HTTP server to serve any dashboard-*.js module from disk.

Changes:

• Add dashboard-dsp.js as the single-source DSP implementation and introduce src/dsp.test.ts to unit-test it under Node.
• Split dashboard.js into feature modules (dashboard-scope.js, dashboard-cheats.js, dashboard-curves.js, dashboard-seq.js) and load them from dashboard.html.
• Update the dashboard server route to serve dashboard.js and dashboard-*.js, and broaden the test runner to execute all compiled *.test.js files.

Reviewed changes

Copilot reviewed 10 out of 11 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
mcp/src/dsp.test.ts	Adds Node unit tests that import the browser DSP script for side effects and validate stabilize/resample/audioPhase.
mcp/dashboard-dsp.js	New single-source DSP implementation shared by browser and Node tests via globalThis.AbxDsp.
mcp/dashboard-dsp.d.ts	Adds a type contract for the global AbxDsp surface.
mcp/src/dashboard.ts	Expands static serving to include dashboard-*.js modules.
mcp/package.json	Updates npm test to run all compiled test files.
mcp/dashboard.html	Loads DSP + core dashboard + feature modules as separate scripts.
mcp/dashboard.js	Shrinks core dashboard shell and delegates features to AbxScope/AbxCheats/AbxCurves/AbxSeq.
mcp/dashboard-scope.js	Moves per-channel scope rendering and its rAF loop into a dedicated module.
mcp/dashboard-cheats.js	Moves cheat sheet + sample browser into a dedicated module.
mcp/dashboard-curves.js	Moves modulation curves UI and paint listeners into a dedicated module.
mcp/dashboard-seq.js	Moves sequencer/pattern/song logic and grid input handlers into a dedicated module.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.